Recent Example Projects

Liquid-S is a Swiss company specialized in Electronic Document Management and Quality Assurance for business processes. The focus is on scanning paper archives and converting them into searchable digital PDF documents enriched with metadata. In many projects, the company processes large volumes of historical documents for enterprise clients.

A key challenge arises when traditional OCR systems fail on complex or low-quality documents. These failures typically occur in cases of complex layouts such as tables, multi-column structures, or text blocks spread across multiple pages. They also occur when document quality is significantly degraded due to repeated copying or physical aging. In some cases, such problematic documents account for a substantial share of entire archives.

To address this challenge, Liquid-S engaged t2b to explore advanced AI-based alternatives. T2b conducted an initial analysis and developed a prototype based on the Thalabus™ platform. The solution combined AI-driven image analysis with large language models for the semantic interpretation of document content. This enabled the system to understand context rather than relying solely on character-level recognition.

The prototype achieved a success rate of 100 percent on the tested dataset. It significantly outperformed traditional OCR methods, which are prone to misinterpretations of visually similar characters. Another challenge in the collaboration was the extraction of metadata from complex structured PDF documents. Conventional extraction tools struggled to identify meaningful text blocks in irregular layouts.

T2b extended the solution using an advanced PDF extraction library combined with LLM-based logic. This approach achieved a success rate of 99 percent in metadata extraction tasks. In addition, the solution improves the quality of scanned images through preprocessing steps such as noise reduction and alignment correction, supports handwriting recognition, and reconstructs complex page structures including tables and embedded elements. Confidentiality is ensured as processing is carried out entirely on local models without any external data transfer.

An important additional aspect of the solution is that Liquid-S operates the Thalabus™-based solution in a GxP-validated environment. This ensures that all AI-driven document processing and metadata workflows comply with strict regulatory and auditable requirements. The approach therefore combines powerful AI capabilities from t2b with full traceability, validation, and compliance for regulated industries.

Based on these results, Liquid-S decided to move the solution into production together with t2b on the Thalabus™ platform. Both companies continue to work closely together to continuously improve the solution and further optimize the models, with the goal of steadily increasing efficiency and performance over time.